Beyond 40 GHz:Chips to be tested, Instruments to measure them

Mark RodwellUniversity of California, Santa Barbara

rodwell@ece.ucsb.edu 805-893-3244, 805-893-3262 fax

>40 GHz Measurements: Why now ?

Very high frequency instruments have existed for some time.

Emerging applications → increased need for instruments40 Gb/s (40-50 Gb line rate) optical fiber transmission60 GHz wireless LANsrates and bandwidths will get still higher

Typical circuit / signal parameters~40-60 GHz circuit bandwidths

40-50 GHz digital clock rates~5-8 ps pulse rise times

~150 GHz transistor cutoff frequencies in medium-scale (2000-transistor) ICs

Instruments neededsampling oscilloscopes (waveform measurements)network analyzer (circuit response, transistor characterization)

Problems facedsampling oscilloscope: timebase stability, connectors, cable loss calibrationnetwork analyzer: cost-effective hardware, precise on-wafer calibration

types of measurements and problems

High-Frequency Measurements: Network Analysis

Circuit frequency response

No extrapolation

Acceptable errors (roughly)~ 0.25 dB in S2130 dB directivity

0

2

4

6

8

10

140 150 160 170 180 190 200

S 21 (d

B)

Frequency (GHz)

Device characterization

S-parameters convertedto Yij and Zij

Frequency variation allowsdevice parameter extraction.

Calibration must be very precise

S22

S12x10S21/10

S11

Ccb,x = 7.1 fF

Rbc = 25 kΩΩΩΩ

Ccb,i = 2.3 fFRbb = 23 ΩΩΩΩ

Cje

Cdiff rbeVbe

rce==250k ΩΩΩΩ

Cpoly= 1.5 fFRex=4.23 ΩΩΩΩ Cout=1 fF

gmVbeexp[-jωωωω(0.23ps)]

rbe=112.5 ΩΩΩΩ, Cje = 47.4 fFCdiff = gm ττττf , ττττf = 0.8866 ps

gm = Ic/VT= 0.239

Base

Emitter

Collector

High-Frequency Measurements: Waveform Measurements

Circuit pulse responseCircuit or system modulation responseFunctioning system

Measurements in 50 Ohm system(Internal node testing not feasible)

Waveforms may berepetitive / periodictransient single-shotrandom data (eyes)

Wideband Optical Transceiver

clockPLL

AD DMUX

O/E, E/O interfaces

MUX

Network Analysis: system-level

microwave network analysis

Block Diagram: microwave network analysis

sampling oscilloscopes

Synthesizers as sampling scope timebase

Simple to implement ( we use this at UCSB)Demands that instruments can control signal stimulusExtremely good timebase stabilityAcceptable for eye patterns, hard for data patterns.

synthesizerdriving display

synthesizerdriving sampler

circuit under test

displayV

t

sampler

synthesizerdriving DUT

LOfIFLORF fNff +=

IFf

IFf

Common triggered-timebase sampling scope

recognizetriggerevent

delay(variable)

delay rampgenerator

displayV

t

sampler

Good: trigger on aperiodic repetitive signalBad: no band limiting in triggering → trigger jitter

PLL

displayV

sampler

VCOtriggersignal frequency

offset

input

PLL as sampling scope timebase ?

PLL is narrowband filter: noise suppression of trigger signal→ less jitterChallenge: maintaining LO frequency within acceptable design rangeAlternatives:

∆−Σ frequency synthesis for frequency offsetDirect digital frequency synthesis timebase control

LOf

NfRF /

IFf

LORF ffNf =∆+/

f∆

RFf

RFf

Cable losses with sampling scopes

Present sampling oscilloscopesdo not provide calibration correction of cable + connector losses

significant ambiguity in waveform measurement, particularly for on-wafer measurements*

*skin losses of wafer probes are high

0

0.2

0.4

0.6

0.8

1

0 5 10 15 20 25 30

Skin Effec t Step Res po ns e

resp

onse

to 1

V in

put s

tep

func

tion

t/τ (norma lized time)

s till ha sn't reached 0.9 Volts !

3 foot V-cable

0 foot cable

Calibration in sampling oscilloscopes ?

Pulse response distortion due to cable loss is becoming major measurement limit.

Network analysis removes such artifacts by calibration

Can NWA calibration be extended to sampling oscilloscopes ?

sampling bridgesand harmonic mixers

Equivalence of sampling and harmonic mixing

Frequency

spectrum of sampling pulse train

Frequency

spectrum of input signal

Frequency

spectrum of IF (sampled) signal

Sampling circuits are one type of high-order harmonic mixer.Sampling circuits used in oscilloscopes and network analyzers

LOfLOf2 LONf

IFLORF fNff +=

IFf

Harmonic-mixing order in network analysis

range dynamicbest hardware expensive

mixing lfundamenta with analysisNetwork

bridge samplingor pair diode bemay mixer harmonic figure noisebetter

tuningLO more LO,higher :hardware expensive moderately orders harmonic low with analysisNetwork

figure noisehigh todue range dynamic degraded tuningLO low frequency, lowat LO :hardware einexpensiv

:orders harmonichigh low with circuits Sampling

n)descriptiodomain (time on) / timeoff (time n)descriptiodomain (frequency conversion oforder harmonic

Circuit Sampling of figure Noise

≥≥

FF

Diode Sampling Bridges

Used in sampling oscilloscopes and network analyzers

Vin VoutVin Vout

)2(2

2.2

inputat risetime

diodeo

RC CZT

RC

=

Schottky diodes are readily made with << 5 fF junction capacitance and » 2 THz R-C cutoff frequencies.

The primary bandwidth limitation of sampling circuits: duration of the strobe pulse used to gate the diodes.

Strobe pulses generated using either Silicon step-recovery diodes, NLTLs, or transistor limiting amplifiers

Risetime / pulse width limits to SRDs

ps 30-20 are devices available-commonlyBest ePerformanc Typical

widthdepletion ofps/micron 10 as thisestimates (1969) Moll regiondepletion in arises collapsecarrier finalin which Time

TimeDiffision Carrier

eCapacitancDepletion

SRDoCZ=τ

NLTL technology

GaAs Schottky diode ICs for mm-wave Instruments

GaAs Nonlinear Transmission Line ICs:0.5 ps pulse generators & DC-725 GHz sampling circuits

0

2

4

6

8

10

0 1 2 3 4 5 6

Cut

off F

requ

ency

, TH

z

Surface doping, N0 , x 1017/cm3

0.5 µm design rules

1 µm design rules

2 µm design rules

Semiconductor Technology:scaled THz Schottky varactor diodes

UCSB, Stanford, Hewlett-Packard 1985-1995

NLTL technologies can cheaply address emerging needs for 100 GHz instruments. Connector and timebase difficulties will dominate cost and accuracy

NLTL Structure and Equivalent Circuit

SPICE Simulation of Shock Formation

Limits to NLTL Shock-Wave Transition Time

Periodic-Network (Bragg) FrequencyThe periodic structure results in a sharp filter cutoff inversely proportionalto the diode spacing. Within lithographic limits, this can easily be 1-2 THz.

Diode Cutoff FrequencyThe fundamental limit of the technology.Falltime limited to

5 THz diode cutoff frequency: 0.28 ps Shock wave

THzps 14.0, •=diodecfall fT

τ2

τ

τ2τ

NLTL-strobed sampling circuit

NLTL & Sampling Bridge, M. Case ~1992

0.1 um Schottky

~700 GHz NLTL-sampler

NLTL using low-loss "air CPW"

~700 GHz NLTL-sampler

measurement of NLTL with NLTL-gated sampling IC

-4.0

-3.0

-2.0

-1.0

0.0

1.0

NLT

L ou

tput

, m

easu

red

by sa

mpl

ing

circ

uit (

Vol

ts)

-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5

0%

100%

Time (ps)

0.68 ps measured10%-90% falltime;0.48 ps deconvolved NLTL falltime,725 GHz deconvolved sampler bandwidth

Aggressive sampling IC design with 1 um diode geometrieslow-loss elevated coplanar waveguide in the NLTLs

measurement of NLTL with NLTL-gated sampling IC

Very simple sampling IC design using 2-3 um process minimum feature sizeSampling bridge bandwidth is approximately 275 GHzDC-110 GHz instruments can be realized using simple, low-cost ICs

Prospective for use of NLTLs in InstrumentsNLTLs have been used commercially since early 1990's HP/Agilent 50 GHz sampling oscilloscopesMicrowave transition analyzer45 MHz -- 50 GHz 8510 network analyzer ?Recent emergence of higher-frequency markets now driving instruments

Less Expensive 45 MHz -110 GHz network analysisPresent systems frequency-combine waveguide-banded systems.

These are accurate but expensive.Reduced-cost instrument could use NLTL-driven sampler for down conversion

NLTL-based sampler can easily down convert DC-200 GHzUse DC-10 GHz synthesized source for LOMixing harmonic order <11 in DC-110 GHz bandwidth: good dynamic rangemain design challenge: LO drive interface to NLTL input

Wider-bandwidth sampling oscilloscopes Present instruments are DC-65 GHz, some are NLTL-based.It is easy to build NLTL-gated samplers far faster than this. Practical limits to DC-110 GHz oscilloscope development are:

timebase stability (eliminating trigger jitter)connector bandwidth limit (110 GHz connectors are fragile)correction of connection (cable, wafer probe) frequency response by calibration.

High FrequencyNetwork Analysis

Active Probes for On-wafer mm-wave network analysis

Sample r/ha rmonic

mixer

Sample r/ha rmonic

mixer

RF source

Sample r/ha rmonic

mixer

Mult iplier

LO sourceHP

8510

Sample r/ha rmonic

mixer

Mult iplier

DUT

Chip

Chip

IF

Couple rLow passfilt e r

St imulus signa l33-50 GHz,

12 dBm

Couple r Couple r

St robe signa l16.5-25 GHz -⌠f,

20 dBm

DUTBias

NLTL

Chip

IF(t o HP8510)

RF Sampler

NLTL

mm-wave interface ICsystem

IC implementation of samplers, multipliers, & couplers allowed for easy system demonstration

→ proof-of-principle demonstration

Close proximity of all components on IC lead to crosstalk, degraded dynamic range.

→ less accurate than needed for real instrument

O WohlegemuthR. Yu

Fraunhofer / UCSB 70-220 GHz Network AnalyzerO WohlegemuthR. Yu

O. Wohlgemuth et al IEEE Transactions on Microwave Theory and Techniques, Vol. 47, No. 12, December.1999

Chip Flexible (GGB)

probe-tip Buffer

amplifier

Fraunhofer / UCSB 70-220 GHz Network Analyzer

80 100 120 140 160 180 200 220-40

-30

-20

-10

0

Mag

nitu

de S

11 [d

B]

Frequency [GHz]

Measurement of S11 of a 900 µm line.

220 GHz

fstart = 70.0 GHz fstop = 230 GHz

Measurement of S21 of a 900 µm line.

O. Wohlgemuth et al IEEE Transactions on Microwave Theory and Techniques, Vol. 47, No. 12, December.1999

O WohlegemuthR. Yu

80 100 120 140 160 180 200 220-5

0

5

10

15

20

25

Dire

ctiv

ity [d

B]

Frequency [GHz]

Measured raw directivity of the ac-tive probe

80 100 120 140 160 180 200 220-60

-55

-50

-45

-40

-35

-30

-25

-20

Max

imum

pow

er [d

Bm]

Frequency [GHz]

Measured maximum power at the IF-ports.

Precision on-wafer > 40 GHznetwork analysis

Network analysis above 40 GHz: commercial toolsAgilent, Wiltron:

RF→ 50 (65) GHz in coax, sampler-basedhigher bands using waveguide and harmonic mixersmultiplexed together for single-sweep measurements

Oleson Microwave Labs. frequency extenders for Agilent, Wiltron140-220 GHz and 220-330 GHz

credits also to the JPL group (T. Gaier et al)

Instruments

ProbesProbes with coaxial connectorsDC-110 GHz, GGB and Cascade

Waveguide coupled probes to 110 GHz (Cascade)to 220 GHz (GGB)to 330 GHz (from GGB soon ?)

On-wafer mm-wave Network Analysis at UCSBApplications:Measurements of transistor amplifiers to 220 GHzPrecise characterization of transistors (power gains, parameter extraction)

45 MHz-50 (40) GHz: Agilent 8510 NWA, coaxial cables and probes

75-110 GHz: Agilent 8510 NWA, waveguide, GGB waveguide-coupled probes

140-220 GHz: Oleson frequency extenders, waveguide, GGB waveguide-coupled probes

Key features for good measurements

on-wafer LRL microstrip calibration standards with offset reference planes

waveguide instrument-probe connections: less loss, less phase drift.

higher band instruments use low-order mixers → better dynamic range

Loss of Coaxial Cable

0.01

0.1

1

10

1 10 100 1000

Atte

nuat

ion,

dB/

met

er

Frequency, GHz

10 mm cable:0.7 dB/m at10 GHz cutoff

3 mm cable:3.1 dB/m at33 GHz cutoff

1 mm cable:14 dB/m at 100 GHz cutoff

1.2=rε 4103)tan( −⋅=δ

( ) 12/1)/2( requiresn propagatio mode-Single −− +⋅≤ outerinnerr DDcf επ2/32/1 Loss / lossSkin fDf skininnerskin ∝→∝ αα

Why waveguide ?

Loss is much lower than for coax(4.5-6 dB / meter in W-band)

> 110 GHz connectors available

No phase drift from Teflon mechanical creep.

HP8510C VNA, Oleson Microwave Lab mm-wave Extenders

GGB Industries coplanar wafer probes

connection via short length of WR-5 waveguide

Internal bias Tees in probes for biasing active devices

75-110 GHz set-up is similar

140-220 GHz On-Wafer Network Analysis

UCSB 140-220 GHz VNA Measurement Set-up

Insertion Loss of Measurement Set-up

<14 dB total attenuation

measurement includes:3 of waveguide connections2 on-wafer probes, through microstrip line (460 um)

Application: Characterizing mm-wave bipolar transistors (HBTs)

Electron beam lithography used to define submicron emitters and collectors

Minimum feature sizes⇒ 0.2 µm emitter stripe widths ⇒ 0.3 µm collector stripe widths

Improved collector-to-emitter alignment using local alignment marks

Aggressive scaling of transistor dimensions predicts progressive improvement of fmax

As we scale HBT to <0.4 um, fmax keeps increasing,measurements become very difficult

0.3 µµµµm Emitter before polyimide planarization

Submicron Collector Stripes(typical: 0.7 um collector)

Miguel Urteaga

Stabilize transistor and simultaneously match input and output of device

Approximate value for hybrid-πmodel

To first order MSG does not depend on fτ or Rbb

Simultaneously match input and output of device

K = Rollet stability factor

How do we measure fmax?

( )1KKSS 2

12

21 −−=MAG

+

≈==

cexcb

12

21

12

21

qIkTRωC

1YY

SS

MSG

For Hybrid- ππππmodel, MSG rolls off at 10 dB/decade, MAG has no fixed slopeCANNOT be used to accurately extrapolate fmax

Maximum Available Gain

Transistor must be unconditionally stable or MAG does not exist

Maximum Stable Gain

ge ne ra tor

los s le s sm a tchin gne twork

R ge n

Vge n

los s le s sma tch ingne twork

R L

loa d

ge ne ra tor

los s le s sm a tchin gne twork

R ge n

Vge n

los s le s sma tch ingne twork

R L

loa d

re s is tivelos s

(s ta b iliz -a tio n)

Miguel Urteaga

Use lossless reactive feedback to cancel device feedback and stabilize the device, then match input/output.

( )12212211

21221

GGGG4YY−

−=U

Unilateral Power GainMasons Unilateral Power Gain

0

5

10

15

20

25

30

35

40

1 10 100

Gai

ns, d

B

Frequency, GHz

MAG/MSGcommon base

U: all 3

MAG/MSGcommon collector

MAG/MSGcommon emitter

For Hybrid- ππππmodel, U rolls off at 20 dB/decade

ALL Power Gains must be unity at fmax

U is not changed by pad reactances

g e n e ra to r

lo s s le s sm a tc h in gn e tw o rk

R g e n

Vg e n

lo s s le s sm a tc h in gn e tw o rk

R L

lo a d

s e rie sfe e d b a c k

s h u n tfe e d b a c k

Miguel Urteaga

On-wafer NWA:

calibration problems

0

5

10

15

20

25

30

35

1 10 100Fre que ncy, GHz

MSG

h21

Ma son'sGa in, U

Submicron HBTs have very low Ccb (< 3 fF) Characterization requires accurate measure of very small S12 Standard 12-term VNA calibrations do not correct S12 background error due to probe-to-probe coupling

SolutionEmbed transistors in sufficient length of transmission line to reduce coupling

Place calibration reference planes at transistor terminals

Line-Reflect-Line CalibrationStandards easily realized on-wafer

Does not require accurate characterization of reflect standards

Characteristics of Line Standards are well controlled in transferred-substrate microstrip wiring environment

Accurate Transistor Measurements Are Not Easy

Transistor in Embedded in LRL Test Structure

230 µm 230 µm

Corrupted 75-110 GHz measurements due toexcessive probe-to-probe coupling

Miguel Urteaga

Line-reflect-line on-wafer cal. standards

Lo Lo

Lo Lo

Lo Lo

Lo+Lo

Lo+560 µm+Lo

Lo+1275 µm+Lo20-60 GHz LINE

75-110 GHz LINE

THROUGH LINE

SHORT

OPEN (reflect)

DUT

75-110 GHz Calibration standards

20-60 GHz Calibration standards

Calibration verification

Device under test

V= 2.04 x 108 m/s(εr = 2.7)

Note that calibration is to line Zo : line Zo is complex at lower frequencies, and must be determined

On-wafer transmission-line wiring environment: impact on LRL calibration

8.3=rε

13=rε

13=rε

m 75 µ=h

m 5 µ=h

m 500 µ=h

Thin-film microstrip

precise LRL calibration

microstrip

might be OK: watch for substrate modesproblem with substrate modesthin wafer ? absorber ?

CPW

140 150 160 170 180 190 200 210 220

freq, GHz

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0.25

0.30

freq (75.00GHz to 110.0GHz)

How good is the calibration ?

freq (140.0GHz to 220.0GHz)

S11 of throughAbout 40 dB

140-220 GHz calibration looks OK75-110 GHz calibration looks Great

S11 of openAbout 0.1 dB / 3o error

dBS21 of through line is off by less than 0.05 dB

S11 of openS11 of short S11 of through

75 80 85 90 95 100 105 110

freq, GHz

-70

-65

-60

-55

-50

-45

-40

Probe-Probe couplingis better than 45 dB

Miguel Urteaga

Measurement of Thru Line after Calibration

S11, S22 (dB)

Magnitude S21 (dB) Phase S21 (degrees)

Miguel Urteaga

Measurement of Line Standard after Calibration

Magnitude S21 (dB) Phase S21 (degrees)

Miguel Urteaga

Measurement after Calibration

Open Standard Line Standard with Open Termination at Second Port

Miguel Urteaga

On-wafer NWA:

results with good LRL calibration

characterization results, DC-40 and W-band

Measurements are smoothresonance-freeconsistent across bandsconsistent with known R's and C's

0

10

20

30

40

1 10 100 1000

Gai

ns (d

B)

Frequency (GHz)

U

h21 462

395

343

139

S22

S12x10S21/10

S11

Ccb,x = 7.1 fF

Rbc = 25 kΩΩΩΩ

Ccb,i = 2.3 fFRbb = 23 ΩΩΩΩ

Cje

Cdiff rbeVbe

rce==250k ΩΩΩΩ

Cpoly= 1.5 fFRex=4.23 ΩΩΩΩ Cout=1 fF

gmVbeexp[-jωωωω(0.23ps)]

rbe=112.5 ΩΩΩΩ, Cje = 47.4 fFCdiff = gm ττττf , ττττf = 0.8866 ps

gm = Ic/VT= 0.239

Base

Emitter

Collector

Sangmin Lee

S11

S22

-6 -4 -2 0 2 4 6

freq (6.000GHz to 45.00GHz)

S12*20

S21

Measurements of Wideband HBTs to 220 GHz

Urteaga et al, 2001 Device Research Conference, June, Notre Dame, Illinois

0

10

20

30

40

10 100 1000

Tran

sist

or G

ains

, dB

Frequency, GHz

U

UMSG/MAG

H21

unbounded U

Collector: 3000 A thickness , 1016/cm3 doping

Collector pulse doping: 50 A thickness 1017/cm3 doping, 250 A from base

Vce

= 1.1 V, Ic=5 mA

0.3 µm x 18 µm emitter, 0.7 µm x 18.6 µm collector,

0 100

2 10-15

4 10-15

6 10-15

8 10-15

1 10-14

0 50 100 150 200

Ic = 1 mAIc = 2 mAIc = 3 mAIc = 4 mAIc = 5 mA

Cc

(Far

ads)

Frequency (GHz)

Vce

=1.1 V

Miguel Urteaga

150 160 170 180 190 200 210140 220

-2

0

2

4

6

-4

8

Frequency, GHz

S2

1, d

B

150 160 170 180 190 200 210140 220

-16

-12

-8

-4

-20

0

Frequency, GHz

S1

1, S

22

, dB

S11

S22

S21

174 GHz Single-HBT Amplifier UCSB

freq (140.0GHz to 220.0GHz)

S11

S22

S21

Measured (solid) and modeled (circle) S-parameters of matching networktest structure

Miguel Urteaga

m2f req=333.0GHzdB(baseline..S(2,1))=0.000

m2

Poorer quality of an on-wafer LRL calibration using CPW

H21

H21

U

U

Mattias Dahlstrom

High Frequency Instruments

Needs:100 GHz sampling oscilloscopes for 40 Gb fiber transmission, Accurate and affordable 60 GHz (100 GHz ?) network analyzers

Easy to Address: sampling (harmonic down conversion) is easy and cheap over DC-200+ GHzother problems are relevant

Sampling Oscilloscopestimebase stability and flexibility in triggering: conflicting requirements !better time bases: 3-synthesizer, PLL, or DDFSchoose timebase appropriate for applicationcable losses are major source of errornetwork-analyzer-like calibration procedures should be developed

Network Analysiscombined accuracy, frequency coverage, and costgood solution (?): moderate-order harmonic conversion with sampler for 40-110 GHzbetter calibration methods needed for testing > 300 GHz ft and fmax transistors